Matthew Schwartz

Statistical Mechanics, Spring 2019

Lecture 10: Quantum Statistical Mechanics

1 Introduction

So far, we have been considering mostly classical statistical mechanics. In classical mechanics,
~p2
energies are continuous, for example momenta ~p and hence kinetic energy E = 2m can be any real
number. Positions ~q can also take any continuous values in aRvolume V = L3. Thus the number of
states
, the entropy S, the partition function Z, all involve dqdp which is formally innite. We
regulated these innities by articially putting limits
q and
p on the positions and momenta
that are allowed, and we found that physical predictions we made did not depend on
p and
q.
Quantum statistical mechanics will allow us to remove these arbitrary cutos.
In quantum mechanics there are two important changes we must take into account. First,
position, momentum and energy are in general quantized. We cannot specify the position and
~
momentum independently, since they are constrained by the uncertainty principle
p
q 6 2 . For
2
example, the momenta modes of a particle in a box of size L are quantized as pn = L ~n, with n an
integer. There is no extra degree of freedom associated with position. When summing over states,
Lp
we sum over n = 2~n only, not over q. However, in the continuum limit we can replace the sum
R
over n with an integral, and suggestively write L = dq. That is, the quantum sum gets replaced
by an integral as
X Z L
Z Z
dqdp
! dn = dp = (1)
2~ h
n

Recall that in classical statistical mechanics when we rst computed

in the microcanonical
ensemble, we had to arbitrarily break up position and momentum into sizes
q and
p. Eq.
(1) justies the replacement
p
q ! h that we previously inserted without proof in the Sackur-
Tetrode equation and elsewhere.1 Note that the integral measure has a factor of h not ~. We'll do
conversions between sums and integrals like in Eq. (1) in more detail in deriving the density of
states over the next several lectures.
Quantization is important when we are not in the continuum limit. One regime where quantum
eects are clearly important is when the temperature is of order the lowest energy of the system,
kBT  "0. For example, classically, each vibrational mode of a molecule contributes NkB to the heat
capacity, but at low temperature, the measured heat capacity does not show this contribution. In
quantum statistical mechanics, the vibrational contribution to the heat capacity is cut o at tem-
peratures below kBT . ~!0, with !0 the vibrational frequency, in agreement with measurements.
We discussed this eect in Lecture 7.
The second quantum eect we need to account for has to do with identical particles. In quantum
mechanics, identical particles of half-integer spin, like the electron, can never occupy the same state,
by the Pauli exclusion principle. These particles obey Fermi-Dirac statistics. Identical particles
of integer spin, like the photon, can occupy the same state but the overall wavefunction must be
symmetric in the interchange of the two particles. Integer-spin particles obey Bose-Einstein
statistics.

1. Eq. (1) makes the appearance of dq seem like a trick. This is an artifact of using momentum eigenstates for our
basis. If we used something symmetric in p and q like coherent states, the product dpdq would appear more naturally.

1
2 Section 2

We discussed indistinguishability before in Lecture 6, in the context of the second law of

thermodynamics and the Gibbs paradox. In that lecture, we found that we could decide if we
want to treat particles as distinguishable or indistinguishable. If we want to treat the particles
as distinguishable, then we must include the entropy increase from measuring the identity of
all the particles to avoid a conict with the second law of thermodynamics. Alternatively, if
we never plan on actually distinguishing them, we can treat them as indistinguishable. We do
1
this by adding a factor of N ! to the number of states
, i.e. instead of
 V N we take

1
N!
V N ; then there is automatically no conict with the second law of thermodynamics. This
kind of classical indistinguishable-particle statistics, with the N ! included, is known as Maxwell-
Boltzmann statistics. Quantum identical particles is a stronger requirement, since it means
the multiparticle wavefunction must be totally symmetric or totally antisymmetric. In a classical
system, the states are continuous, so there is exactly zero chance of two particles being in the same
state. Thus, the dierence among Fermi-Dirac, Bose-Einstein and Maxwell-Boltzmann statistics
arises entirely from situations were a single state has a nonzero change of being multiply occupied.

2 Bosons and fermions

1
The wavefunction (x; s) of an electron depends on the coordinate x and the spin s =  2 . When
there are two electrons, the wavefunction depends on two coordinates and two spins. We can write
it as (x1; s1; x2; s2). Thus j (x1; s1; x2; s2)j2 gives the probability of nding one electron at x1 with
spin s1 and one electron at x2 with spin s2. Every electron is identical and therefore2
j (x1; s1; x2; s2)j2 = j (x2; s2; x1; s1)j2 (2)
If the modulus of two complex numbers is the same, they can only dier by a phase. We thus have
(x1; s1; x2; x2) =  (x2; s2; x1; s1) (3)
for some phase  = ei with j j = 1. If we swap the particles back, we then nd
(x1; s1; x2; s2) =  (x2; s2; x1; s1) =  2 (x1; s1; x2; s2) (4)
So that  2 = 1 and  = 1. We call particles with  = 1 bosons and those with  = ¡1 fermions.
1 3
Particles come with dierent spins s = 0; 2 ; 1; 2 ;


. A beautiful result from quantum eld theory
is that particles with integer spins, s = 0; 1; 2;


are bosons and particles with half-integrer spins
1 3
s = 2 ; 2 ;


are fermions. This correspondance is known as the spin-statistics theorem.
As an aside, I'll give a quick explanation of where the spin-statistics theorem comes from. First
we need to know what spin means. The spin determines how the state j i changes under rotations.
For example, say a particle is localized at x = 0, so (x ~ ) = hxj i =  3(x), and has spin pointing in
the z direction, so jz = s. Now rotate it around the x axis by an angle . Doing so means it will
pick up a phase determined by the spin
j i ! eis j i (5)
In other words, the spin s is the coecient of  in the phase. For example, write the electric eld
vector E~ as a complex number z = Ex + iEy. Then under a rotation by angle , z ! eiz, so s = 1
according to Eq. (5). This tells us that photons (the quanta of the electric eld) have spin 1.
Under a 360 rotation ( = 2), we are back to where we started. But that doesn't mean that
j i ! j i. Indeed, we see directly that if s is a half-integer, then j i = ¡j i while if s is an integer
then j i ! j i. It may seem a little weird that we could rotate something around 360 and not
get to where we started. It turns out this is possible due to a funny mathematical property of the
set of all possible rotations, the rotation group SO(3): it is not topologically simply connected. To
see this, try holding a glass in your hand and rotate the glass around 360. You will nd your arm
all twisted up. If you rotate by another 360 you will be back to where you started.
2. Electrons are identical by denition, since if every electron were not identical, we could give the dierent
electrons dierent names. For example, there's a particle called the muon which is similar to the electron but
heavier. Thus the electron/muon wavefunction will have (x1; x2) = / (x2; x1). Unless there are an innite number
of possible quantum numbers that we can use to tell all electrons apart, some of them will be identical and satisfy
j (x1; s1; x2; x2)j2 = j (x2; s2; x1; s1)j2. The identical particles for which quantum statistics applies are, by denition,
these identical ones.
Bosons and fermions 3

Now consider a two particle state, (x1; "; x2; ") with s = " denoting spin up in the z direction.
Now we rotate by 180 around the center between x1 and x2, this interchanges the particles, but
adds a phase eis for particle 1 and another phase eis for particle 2. So the interchange adds a
factor of e2is = 1 total to the wavefunction. This is exactly the factor  in Eq. (3). Thus  = e2is
which is 1 for integer spins and ¡1 for half-integer spins. In other words, integer spin particles are
bosons and half-integer spin particles are fermions.
The main implication of the spin-statistics theorem for statistical mechanics is the Pauli exclu-
sion principle: no two fermions (like electrons) with the same quantum numbers (like spin) can
occupy the same state. To see this, say the wavefunction for an state is 0(x; s). Then if two
particles were in the same state with the same spin then (x2; s2; x1; s1) = 0(x1; s1) 0(x2; s2). This
is symmetric under interchange. So if (x2; s2; x1; s1) = ¡ (x1; s1; x2; s2), as for fermions, then
must vanish. Instead, for fermions we can only have (x2; s2; x1; s1) = 0(x1; s1) 1(x2; s2) with two
dierent single particle wavefunctions 0 and 1. More generally, each identical fermion must be
in a dierent state, so when we pile N fermions into a system, they start lling the energy levels
from the bottom up.

2.1 Three types of statistics

There are 3 types of statistics we will discuss
 In (classical) Maxwell-Boltzmann statistics any particle can be in any state and an overall
1 1 P
factor of N ! for indistinguishability is added ad hoc to the number of states
= N ! k 1 (micro-
1 P
canonical ensemble) or to partition function: Z = N ! k e¡Ek (canonical ensemble).
 In Bose-Einstein statistics multiple particles can occupy the same state.PBut when they do,
there is only one state with the multiple particles in it. No N ! is added: Z = e¡Ek.
PIn Fermi-Dirac statistics, no two particles can occupy the same state. No N ! is added and
Z = e¡Ek.
Let's compare the situations with an example. Suppose there are 2 particles and 3 possible
single-particle states with energies "1, "2 and "3. These energies do not have to be dierent, but
the single-particle states should be distinguished by some quantum numbers, such as dierent
spin or position or vibrational excitation along a dierent axis. For Maxwell-Boltzmann statistics,
1
the particles are treated as distinguishable for counting states, and the N ! is added at the end.
Denoting the two particles A and B, the possible states and canonical partition function (with
 = 1 for simplicity) in this case is:

"1 "2 "3

AB
A B
A B
B A 1 ¡2"1
; ZMB = [e + e¡2"2 + e¡2"3 + 2e¡"1 ¡"2 + 2e¡"1 ¡"3 + 2e¡"2 ¡"3] (6)
AB 2!
A B
B A
B A
AB
1
The 2! is the identical particles factor. Note that we treat the two particles as dierent when
constructing the partition function, then divide by N !:
For Bose-Einstein statistics, the possible states and canonical partition function are
"1 "2 "3
AA
A A
A A ; ZBE = e¡2"1 + e¡2"2 + e¡2"3 + e¡"1 ¡"2 + e¡"1 ¡"3 + e¡"2 ¡"3 (7)
AA
A A
AA
4 Section 3

Note that we treat the particles as identical, so AA = BA = AB is one state. No N ! is necessary.

For Fermi-Dirac statistics, the possible states and canonical partition function are

"1 "2 "3

A A
; ZFD = e¡"1 ¡"2 + e¡"1 ¡"3 + e¡"2 ¡"3 (8)
A A
A A

Note that there are 9 two-particle states for the Maxwell-Boltzmann case, 6 for the boson case
and only 3 for the fermion case. The 3 states in the fermion case are the ones where each particle
has a dierent energy. If there were m possible energy levels instead of 3, then there would be
 
m 1 1
nfermi = = m2 ¡ m (9)
2 2 2

possible fermion states. The number of boson two-particle states would include all the possible
fermion ones plus the m states where both particles have the same energy, so

1 1
nbose = nfermi + m = m2 + m (10)
2 2

The number of classical two-particle states is nclass = m2. To compare to the fermion or boson
cases, note that in the limit that kBT  "i for all the energies, the Boltzmann factors all go to 1
so the partition function Z just counts the number states. Thus the eective number of states in
Maxwell-Boltzmann statistics is
1 1
nMB = ZMB(kBT  "j ) = nclass = m2 (11)
2! 2
Thus we see that for 2 particles when the number of thermally accessible energies m is large, the
number of two-particle states is the same in all three cases.
For N particles and m energy levels, the same analysis applies. The dierence between the
fermion and boson cases is due to the importance of states where more than one particle occupy
the same energy level. The number of such states is always down by at least a factor of m from
the number of states where all the particles have dierent energies. Thus when m gets very large,
so that there are a large number of thermally accessible energy levels, then the distinction between
bosons and fermions starts to vanish. In other words, m  N almost all the microstates will have
all the particles at dierent energies. For any particular set of N energies for the N particles there
would be N ! classical assignments of the particles to the energy levels, but only one quantum
assignment. This explains why the N ! Gibbs factor for identical particles in the Maxwell-Boltzmann
statistics is consistent with quantum statistical mechanics. We will make more quantitative the
limit in which quantum statistics is important in Section 4.

3 Non-interacting gases
Next we compute the probability distributions for the dierent statistics, assuming only that
the particles are non-interacting. We call the non-interacting systems gases, i.e. Bose gas and
Fermi gas. Neglecting interactions allows us to write the total energy as the sum of the individual
particle energies. For photons, neglecting interactions is an excellent approximation since Maxwell's
equations are linear  the electromagnetic elds does not interact with itself. For fermions, like
electrons, even though there are electromagnetic interactions, because fermions like to stay away
from each other (due to the Pauli exclusion principle), the interactions are generally weak. Thus
the non-interacting-gas approximation will be excellent for many bosonic and fermionic systems,
as we will see over the next 5 lectures.
Non-interacting gases 5

Let us index the possible energy levels each particle can be have by i = 1; 2; 3;


. The energy of
n
level i is "i. For example, a particle in a box of size L has quantized momenta, pn = L ~ and energies
2
pn
"n = 2m . In any microstate with a given N and E there will be some number ni of particles at level
P P
i, so ni = N and ni"i = E. It is actually very dicult to work with either the microcanonical
ensemble (xed N and E) or the canonical ensemble (xed N and T ) for quantum statistics. For
example, with 2 particles, the canonical partition function for a Bose gas would be
1 X
X 1
ZBE = e¡("i +"j) (12)
i=1 j=i

You can check that with only 3 energy levels, this reproduces Eq. (7). Note that we must have
j > i in the second sum to avoid double counting the (i; j) and (j ; i) states. For N particles, we
would need N indices and can write
X
ZBE = e¡("i1 +"i2 +


+"iN ) (13)
i1 6i2 6


6iN

Doing the sum this way over N indices is very dicult; even in the simplest cases it is impossible
to do in closed form. For Fermi-Dirac statistics the partition function can be written the same way
with the sum over indices with strict ordering: i1 < i2 <


< iN . It is also very hard to do.
Conveniently, the grand canonical ensemble comes to the rescue. In the grand canonical
ensemble, we sum over N and E, xing instead  and T . The grand partition function is
X
Z= e¡(E ¡N) (14)
N ;E
P P
Since ni = N and ni"i = E the grand partition function can also be written as
Y 1
X
Z= e¡ni("i ¡) (15)
single particle states i ni =0

where ni is the number of particles in state i. To see that this agrees with Eq. (14) consider that
each term in the product over i picks one factor from each term in the sum over ni. Since allP possible
values of "i and ni are included once, we reproduce the sum over all possible N and E with ni = N
P
and ni"i = E. For example, if ni = 0; 1; 2;


are allowed (as in a Bose system), then Eq. (15) means

Z = (1 + e¡"1 + e¡2"1 +


)(1 + e¡"2 + e¡2"2 +


)(1 + e¡"3 + e¡2"3 +


)


(16)

=


+ e2"1 + e2"2 + e2"3 + e"1 +"2 + e"1 +"3 + e"2 +"3 +


(17)

The second line shows all terms with N = 2 and energies "1; "2; "3 and agrees with the explicit
enumeration in Eq. (7). The equivalence of Eq. (15) and Eq. (14) is important. So please make
sure you understand it, working out your own checks as needed.
Eq. (15) can be written more suggestively as
Y
Z= Zi (18)
singleparticlestatesi
with
X
Zi = e¡ni("i ¡ ) (19)
possibleoccupanciesni

This factor Zi is the one-state grand-canonical partition function, i.e. the grand-canonical partition
function for a system with only one single-particle state, although the state can be multiply-
occupied. Eq. (18) means that the full grand partition function is a product of the grand partition
functions for the separate single-particle states: each degree of freedom can be excited indepen-
dently. This is very powerful, since it means the probability of nding ni particles occupying state
i is completely independent of whatever else is happening. This is true at xed  but wouldn't be
true at xed N  try to factor Eq. (7) in this way; it doesn't work.
6 Section 3

The grand canonical ensemble makes it very easy to write an expression for the expected
occupation number of state i:
1X 1 @lnZi
hnii = nie¡ni("i ¡ ) = (20)
Zi  @
ni

We can also express this as

@i
hnii = ¡ (21)
@
where
1
i = ¡ lnZi (22)

is the grand free energy for state i.

3.1 Non-interacting Bose gas

For a Bose system, there can be any number ni = 0; 1; 2;


of bosons in a given state i. The single-
state grand-canonical partition function for the state i with energy "i, from Eq. (19), is then
1
X
Zi = e¡n("i ¡ ) (23)
n=0

This is a simple geometric series: there can be 0 particles, 1 particles, 2 particles, etc in the state,
and so the total amount of energy in the state is an integer multiple of "i. Performing the sum,
we nd
X1
1
Zi = e¡n("i ¡) = ¡("i ¡ )
(24)
n=0
1¡e

The full partition function from Eq. (18) is then

Y Y 1
Z= Zi = ¡("i ¡ )
(25)
i i
1 ¡ e

Note again that the product is over all the dierent states i, with "i the energy of that state.
The grand free energy for each state is
1 1
i = ¡ ln Zi = ln(1 ¡ e¡("i ¡ )) (26)
 
Then the occupation number for each state is
(27)

@i 1
hnii = ¡ = (" ¡ ) =
@ e i
¡1

This is known as the Bose-Einstein distribution.

Note that if "i <  then hnii is negative, which is impossible. This means that  < "i for all "i,
i.e. the chemical potential for any Bose system will be less than all of the energies. Often we set
the ground state energy to zero in which case
  < 0 for bosons
Non-interacting gases 7

Negative chemical potential is consistent with the formula we derived in Lecture 7 for  for a
classical monatomic ideal gas:  = kBT ln n3  ¡0.39 eV < 0 with n the number density and  the
thermal wavelength. We'll use the chemical potential for bosons in Lectures 11 and 12.

3.2 Non-interacting Fermi gas

For fermions, no two particles can occupy the same single-particle state. So the one-state grand-
canonical partition function is simply
X
Zi = e¡("i ¡ )n = 1 + e¡("i ¡ ) (28)
n=0;1

The full grand-canonical partition function is then

Y Y
Z= Zi = [1 + e¡("i ¡ )] (29)
i i

The grand free energy for state i is

1 1
i = ¡ ln Zi = ¡ ln[1 + e¡("i ¡ )] (30)
 

So that the occupation number for state i is

(31)

@i 1
hnii = ¡ = (" ¡ )  f (") =
@ e i
+1

This is known as the Fermi-Dirac distribution or Fermi function, f("). We see that at low
temperature, states with energy greater than  are essentially unoccupied: f (" > )  0 and states
with energies less than  are completely lled.
The chemical potential at T = 0 is called the Fermi energy, "F = (T = 0). At T = 0, all states
below the Fermi energy are lled and the ones above the Fermi energy are empty. Thus the chemical
potential for Fermi systems is positive at low temperature and corresponds to the highest occupied
energy level of the system. We'll use the Fermi energy extensively to discuss fermionic systems in
the quantum regime, in Lectures 13, 14 and 15.
@ S
As the temperature increases, the chemical potential decreases. This follows from @T = ¡ N and
S > 0. Eventually, the chemical potential will drop below all of the energy levels of the system.
That is, it becomes negative, as with a Bose system or a classical gas.

3.3 Maxwell-Boltzmann statistics

Maxwell-Boltzmann statistics are the statistics of classical indistinguishable particles. The parti-
tion function is computed by summing over states with any particles in any state, and the sum is
divided by N ! to account for indistinguishability. Although it is often an excellent approximation,
no physical system actually obeys Maxwell-Boltzmann statistics exactly. Nevertheless, counting
states this way turns out to be very much easier than using Bose or Fermi statistics and gives the
same answer in the continuum/classical limit.
8 Section 3

For Bose or Fermi gases, we were able toQwrite the partition function as the product of partition
functions for each state, as in Eq. (18): Z = i Zi. With Maxwell-Boltzmann statistics, it's not clear
yet if we can do this since we must divide by N ! where N is the total number of particles, not just
the number of particles in state i. To avoid this subtlety, let's instead compute the grand-canonical
partition function directly from the canonical partition function. Indeed, we can always write
X
Z= ZNe N (32)
N

where ZN is the canonical partition function with N particles. We could have tried this for Bose
or Fermi statistics, but for those cases the canonical partition function ZN with N xed is hard to
evaluate.
The canonical partition function ZN with Maxwell-Boltzmann statistics is
1 X
ZN = e¡Ek (33)
N!
distinguishable
N particlemicrostatesk

In this sum the particles are treated as distinguishable and indistinguishability is enforced entirely
through the N ! factor.3
Since each particle is independent and the total energy E is unconstrained, we can pick any
state i for any particle and therefore
1 X X ¡("i1 +


+"iN )
ZN =


e (34)
N!
i1 iN

Here the sum over ij is over the possible states i for particle j. Since the sums are all the same,
we can simplify this as in Eq. (15):
  X 
1 X ¡"i 1 N
ZN = e


e¡"i = Z (35)
N! N! 1
i i

where the single-particle canonical partition function is

X
Z1 = e¡"i (36)
i

The grand partition function for Maxwell-Boltzmann statistics is then

X 1 X  Y
Z= Z1NeN = exp[Z1e] = exp e¡("i ¡ ) = exp[e¡("i ¡ )] (37)
N!
N i i

This has the same form as Eq. (18) after all

Y
Z= Zi (38)
i
with
Zi = exp[e¡("i ¡ )] (39)
In fact, if we write
X 1
Zi = e¡n("i ¡ ) (40)
n!
n

We see that Zi is exactly the grand partition function for a single state using Maxwell-Boltzmann
statistics.
The grand free energy for state i is then
1 1
i = ¡ lnZi = ¡ e¡("¡ ) (41)
 
3. Note that dividing by N ! is a little too much. N ! is the number of permutations when the particles are in
dierent states, such as (A; B; 0) and (B; A; 0) in the table in Eq. (6). But it divides by too much when there is
more than one particle in the same state, like the state (AB; 0; 0) in the top row in the table in Eq. (6); there is
only one state like this, not two. This approximation is valid when N is much larger than the number of accessible
states, as at high temperature or in the classical continuum limit. In such limits, the number of congurations with
more than one particle in a state is negligible.
Quantum and classical regime 9

and the expected number of particles in state i is

@i
hnii = ¡ = e¡("i ¡) = (42)
@

This is known as the Maxwell-Boltzmann distribution function. It looks a lot like the Bose-
Einstein distribution function.
Unlike the Bose-Einstein case, taking  > "i does not give negative occupation numbers. It can
however, give occupation numbers greater than 1. When states are multiply occupied then using
classical statistics is no longer justied and we must use Bose or Fermi statistics. Thus for Maxwell-
Boltzmann statistics, we should have  < "i for all "i. Or, setting the ground state energy "0 = 0 we
should have  < 0. Note that this is consistent with our understanding of  from classical statistical
thermodynamics. For example, in Lecture 7 we saw that for a monatomic ideal gas with "0 = 0
h
then  = kBT ln(n3) with  = p the thermal wavelength. As long as we are at low density
2m kBT
n < ¡3, then  < 0. When n  ¡3 then quantum eects become important and the sign of  will
depend on whether the gas is bosonic or fermionic, i.e. our classical computation of the monatomic
ideal gas is no longer valid. We next discuss this quantum/classical transition in more detail.

4 Quantum and classical regime

Let's summarize the 3 types of statistics:
 Fermi-Dirac statistics applies to particles of half-integer spin like electrons in metals or white-
dwarf stars or half-integer spin nuclei like protons or neutrons. Fermi-Dirac statistics applies not
just to elementary particles, but also to atoms that have an odd number of fermions, such as 3He,
which comprises 2 protons, 1 neutron and 2 electrons. In Fermi-Dirac statistics, no two identical
particles can occupy the same single-particle state.
 Bose-Einstein statistics applies to particles of integer spin, like photons, phonons, vibrational
modes, 4He atoms (=2 neutrons, 2 protons and 2 electron) or 95Rb atoms (35 protons, 60 neutrons,
35 electrons). In Bose-Einstein statistics, any number of bosons can occupy the same state.
 Maxwell-Boltzmann statistics are a kind of phony classical statistics for which calculations
are generally easier than with bosons or fermions. We treat the particles as all distinguishable,
1
then throw in a factor of N ! to the partition function to account for indistinguishability.
The main results of this lecture were the expressions for the partition functions and probability
distributions for the various statistics. We observed that for ideal gases (non-interacting particles)
with any statistics, the grand-canonical partition function can be written as
Y
Z= Zi (43)
singleparticlestatesi

where Zi is the grand-canonical partition function for a single state. For the various statistics we
found
1
Zi = ¡("i ¡ )
(Bose ¡ Einstein) (44)
1¡e
Zi = 1 + e¡("i ¡ ) (Fermi ¡ Dirac) (45)
Zi = exp[e¡("i ¡)] (Maxwell ¡ Boltzmann) (46)
10 Section 4

From these, we deduce the probabilities

 hnii for nding ni particles in a given state i with energy
@ 1
"i. The general formula is hnii = @  lnZi . We found

1
hnii = ("i ¡ )
(Bose ¡ Einstein) (47)
e ¡1
1
hnii = ("i ¡)
(Fermi ¡ Dirac) (48)
e +1
hnii = e¡("i ¡ ) (Maxwell ¡ Boltzmann) (49)
Note that all of these have the form
1
hnii = (50)
e ("i ¡ ) + c
with c = ¡1; 0 or 1.
The classical limit is when the probability two particles being in the same state is irrelevant.
This limit requires that the hnii are all small, hnii  1, which amounts to

e("i ¡)  1 (51)

In this limit, all the statistics gives the same results
hnii  e¡("i ¡ ) (52)

You might be a little bothered by the condition e("i ¡)  1 for the classical limit, since  ! 1
is T ! 0, so this seems like a low temperature limit. It certainly can be a low temperature limit:
taking kBT  " is one way to make sure that the state with energy " is unpopulated. But the
classical limit can be achieved other ways as well. The key point you have to keep in mind is that
while the grand canonical ensemble works at xed , our intuition is for xed N . If N is xed,
then  is a dependent variable and has strong temperature dependence, stronger even than .
For example, we showed that for a classical monatomic ideal gas the number density and chemical
N 1 h
potential are related by V = 3 exp() with  = p the thermal wavelength. So for Maxwell-
2mkBT
Boltzmann statistics at xed N we can write
 3/2
¡("i ¡) N h2 "
¡ i
hnii = e = e kBT (53)
V 2mkBT
This is, of course, just the original Maxwell-Boltzmann distribution we derived back in Lecture
1
3. Note that at large T the exponential goes to 1 and the prefactor 3  3/2 goes to zero. More
T
physically, at high T , more and more states become accessible so it is less and less likely for any
state to be multiply occupied. In addition, at high T particles are moving fast and their de Broglie
wavelengths shrink: they become more like classical particles. So we can have hnii ! 0 either when
T ! 0 at xed  or when T ! 1 at xed N . Another way is simply taking V ! 1 to make more
and more states. There are many ways to get hnii  1.
Quite generally, increasing the temperature at xed N decreases the chemical potential, since
@ S
@T
= ¡ N < 0. For a Bose system,  < "i for all "i so "i ¡  > 0 and grows with temperature. In fact,
S
it grows with temperature faster than  decreases with temperature: roughly " ¡  N T  fT ln T
1
for some f so (" ¡ )  f lnT and hnii  T f at large T . Thus we see more broadly that high
temperature gives classical statistics, as you would expect, despite the fact that a supercial reading
of e¡("i ¡ )  1 makes it seem otherwise.
I suspect a number of you are cursing the chemical potential at this point. Why don't we just
the microcanonical or canonical ensembles, at xed N , rather than the grand canonical ensemble
with its unintuitive ? As I have been saying, it is extremely dicult to use the microcanonical
or canonical ensembles for Bose-Einstein of Fermi-Dirac statistics due to the challenging combina-
torics of counting microstates at xed N . As we will see, having  around is not bad at all once
you get used to it.  is essential to understanding Bose-Einstein condensation (Lecture 12) and
metals (Lectures 13 and 14).
Quantum and classical regime 11

We have seen that many limits (large T , small T , large V ) lead to classical statistics. The
challenge is to nd a regime where the quantum statistics dominate. For quantum statistics to
be relavant, we need hnii / 1. Since the higher energy states are going to be less populated than
the ground state, a reasonable condition is that hn0i  1. Using Eq. (53), the ground state with
"i = 0 has hn0i = 1 when
 3/2
N 2mkBT 1
= = 3 (54)
V h2 
 1/3
V
So we want   N . In other words

 quantum statistics are important when the thermal de Broglie wavelength is of

order or bigger than the interparticle spacing
For example, at room temperature air  1.87  10¡11m while the average interparticle spacing in
 1/3  
V k T 1/3
air is N = BP = 3  10¡9, so particles in air are around a factor of 100 times too far for
quantum eects to be important. If we keep the pressure at 1 atm, we would have to cool air to
T = 0.57K for quantum eects to matter.
N
For a classical monatomic ideal gas  = kBT ln(3n). Thus in the classical regime when 3 V  1
the chemical potential is a large negative number,   0. As temperature is decreased at constant
N and V then  goes up and 3n increases. As 3n gets close to 1 then  ! 0 from below (more
precisely,  ! "0 with "0 the ground state energy, but we usually set "0 = 0). As  ! 0, whether
the system is bosonic or fermionic becomes important. For a bosonic system,  can get closer
and closer to 0, but can never reach it. As we will see in Lecture 12, as  ! 0, bosons accumulate
in the ground state, a process called Bose-Einstein condensation. For a fermionic system,  goes
right through zero and ends up at a positive value the Fermi energy  = "F > 0 at T = 0. Thus the
closeness of  to 0 or equivalently the closeness of 3n to 1 indicates the onset of quantum statistics.